Emotional Lpc Coefficients Based On Gaussian Mixture Model Research And Modeling

With the rapid development of information technology, human dependence on computer is growing. Therefore, human-computer interaction capabilities get more and more attention of researchers. Voice has the largest information capacity in many information carrier, with the highest level of intelligence. Traditional voice processing systems focus on the accuracy of voice only, ignoring the emotional factors contained in the voice signal. Therefore, this article studied on emotional speech modeling of lpc coefficients.According to status of deficiency of study on emotional speech modeling, the paper presents a new method of emotional speech modeling of lpc coefficients. The paper established four kinds of emotional speech database including happy, angry, sad and neutral emotion recorded by Chinese Academy of Sciences; researched acoustic characteristic parameters; got Resonance peak statistical regularity of different emotion; designed and realized the new modeling scheme of emotional lpc coefficient. It used different emotional LPC feature vector, combined with the dynamic time warping technology, EM algorithm and MMSE criterion, finally got lpc coefficient mapping rule function of three kinds of emotional speech to neutral speech; and completed the emotional lpc parameter modeling. Also the paper designed the experimental test plan, calculated spectrum distortion measure between the standard neutral voice lpc coefficient and lpc coefficient which got by mapping function using IS distance. Simulation results show that the emotional speech model can efficiently characterize the different emotional effect on lpc coefficient.The new method of motional speech modeling of lpc coefficients that this paper presented is a new method of emotional speech signal processing field and provides a new idea and solution to the research of the influence for emotional speech synthesis and recognition.